Purpose
RBD is the front end most Ceph clusters serve first, and its failure modes are
quiet. An image created with features the kernel client does not implement will
map fine on one host and refuse on another; a min_size of 1 will look
identical to a correct pool until the day it silently accepts writes it cannot
protect. This checklist is the gate before the first production volume.
Order of work
The list runs in three passes. The durability pass is size-min-size and
vms: how many copies exist and which rule places them, settled first because
they are the items that decide whether data survives. The client pass is
features and caps - what the consumer is capable of mapping and what it
is permitted to reach, checked against the actual client in use rather than
against the default. Kernel RBD and librbd do not support the same feature set,
and the checklist is answered for whichever one the workload runs. The
data-protection pass is snapshots and backup, which are two different
mechanisms and not alternatives to each other.
Before you start
Have the client type and kernel version for each consumer, the durability target for the workload, and the existing cephx identities.
Verification
Ready when every critical item is checked and an image has been mapped, written to, snapshotted and unmapped from a representative client. Mark an item N/A rather than leaving it unchecked when it does not apply to this cluster.
Escalation
A capability set broader than the workload needs is a security finding, not a convenience. Route it through the storage lead.